What is Artificial Intelligence (AI)?
Another attempt at a definition and other ramblings
“What is AI?” you ask.
“I don’t know.”, I reply. “It’s not something I think about often.”
“But don’t you earn a living out of AI?”
“Well — yes.”, I muse. “But that’s like asking a Taxi Driver about an internal combustion engine or continuously variable transmission (CVT). You don’t need to know how a car engine works to earn a living out of a car.”
“A Taxi Driver might not know. But an Automobile Engineer will. Are you saying that you’re more like a Taxi Driver when it comes to AI?”
“No, no”, I reply hastily. “I do think about what AI is — sometimes.”
“So then, what is it?”
Chess
“Well, it’s complicated. Artificial is easy enough to define. Intelligence is harder.”
“Is it? I know what I mean when I say that my ten-year-old niece is intelligent. What’s the difference?
“Well — what do you mean when you say your niece is intelligent?”
“Well — she does intelligent things. She can play Chess, for example.”
“How did she learn to play chess?”
“I taught her”
“How?”
“I first taught her the names of all the pieces. Then, where they go on the board. And finally, what moves the pieces make.”
“Anything else?”
“Well, after a few games, I taught her to “look ahead” for a few moves. Not to move her Bishop, for example, because it would expose her Queen.”
“A computer that plays Chess is the same. It knows the rules, and it can look ahead for the best moves. The most powerful chess computers can look ahead for billions of combinations of moves.”
“What you’re saying is that the Computer is intelligent because it does things that make humans intelligent?”
“Sort of. Some AI experts defined AI this way. As in, “Computers that can do human tasks considered intelligent”.
“Is that what AI is?
“I don’t like that definition because it is “relative” to Human Intelligence. And we don’t define what Human Intelligence is either. It is also subjective. If playing Chess is intelligent, what about playing Tik-Tak-Toe? Not everyone might call a Tik-Tak-Toe playing computer an AI.”
Expanding Data
“Do you have a better definition?”
“Well, we could think about what a Human or a Computer does when it does a supposedly intelligent task?”
“Like “thinking ahead”?”
“Exactly. The intelligence of playing Chess seems to come entirely from thinking ahead. You start with the rules and expand this to millions of combinations of Chess games.”
“And this is intelligent?”
“Yes — at least it is the essence of intelligent.”
“Thinking ahead?”
“Well, thinking ahead is an example of the essence. The essence is taking some data, like the rules of Chess, and expanding it to more data, like millions of combinations of future games, which will tell you what the best next move is.”
“So AI is Machines extending data?”
Cats and Dogs
“That’s the essence. Let me give you another example: Recognizing Cats and Dogs.”
“Like an AI that can tell Cats and Dogs apart?”
“Yes.”
“How do these “Expand data”?”
“Well, you train a model that can tell the difference, using millions of examples of images of Cats and Dogs, labelled CAT or DOG. “Training a model” consists of building a mathematical function that transforms the image into some other representation which can accurately represent either CAT or DOG.”
“But where is data expanded?”
“The model is an expansion of the training data.”
“But is this ability intelligence? A child can learn to recognize a Cat or a Dog after seeing just one or two animals. Why can’t a Machine do the same?”
“I think that’s because the child has mostly learnt to recognize Cats and Dogs before it is born.”
“Before? How is that possible?”
“Well, the “model training process” in humans has happened through millions of years of evolution. Most of the model is complete. Only the very last step is necessary. That is why just a few examples are sufficient. That’s because millions of our ancestors have already seen millions of Cats and Dogs.”
Technology
“So, what drives AI? How come AI has developed so fast in the last couple of decades?”
“It’s because Machines have got better and better at data expansion.”
“How?”
“Take the example of the Chess Computer. The algorithms for “looking ahead” are relatively simple and have been around for years. It’s just as computers have got faster; it’s possible to look ahead for more and more moves.”
“I see. And what about Cats and Dogs? Are Object Recognizing AIs also running faster algorithms?”
“Yes. We can train models quicker. A model that previously took many years now takes a few minutes. But it’s not just faster models.”
“What else?”
“There’s also been an explosion of data. Even if we could train models fast, they would not work without millions of Cat and Dog photos. And if platforms like Facebook and YouTube didn’t have these photos and videos, it would not have been possible to train models.”
“So how come YouTube have so many Cat videos?”
“Well — a combination of many things. Advances in technology have led to bigger data stores, faster networks, and ubiquitous devices like Smartphones, all of which are also much cheaper. As a result, products like YouTube and Facebook, which encourage humans to post Cats and Dogs, also become possible.”
“So, it is all about technology?”
“Yes.”
Human Intelligence
“So tell me something: Do you think computers are more intelligent than humans? Will they ever be?”
“I don’t know.”
“But why? How can my 1.5 Kg brain be more intelligent than millions of racks of computers combined?”
“Why did you mention your brain?”
“Isn’t that where all the intelligence happens?”
“How?”
“Isn’t that where all the processing happens? And where we store our memory?”
“Well, we don’t know that for sure. We know that humans can use memory sources outside our bodies, like taking notes in a notebook. Hence, the whole universe could be a sort of memory, and I don’t know what fraction of it we use and what we don’t use. The same applies to processing. We don’t know where exactly “processing” happens. Most of our processing is subconscious, and we don’t even know where the conscious part happens.”
“Will machines start using the universe as a memory? They seem to be doing that with things like the “Internet of Things”.”
“Yes.”
“So why do you think machines will never be as intelligent as humans?”
“I didn’t say that. I said — I don’t know.”
“Why don’t you know?”
“Well, because I don’t think that I or anyone else knows exactly how intelligent humans can be. And if we don’t know that, how can we say that something else (like a machine) will be more so?”
“So you think humans might be capable of being more intelligent that they are? To use your definition, they might be able to “expand data” more than they are already doing?”
“Yes.”
“How?”
“I don’t know. But we should be more interested and enthusiastic about questions like this.”
“Like we are interested and enthusiastic about AI?”
“More so!”
